Received October 31, 2020, accepted November 24, 2020, date of publication November 27, 2020,

date of current version December 11, 2020.

Digital Object Identifier 10.1109/ACCESS.2020.3041084

Detection of Financial Opportunities

in Micro-Blogging Data With a
Stacked Classification System
FRANCISCO DE ARRIBA-PÉREZ, SILVIA GARCÍA-MÉNDEZ ,
JOSÉ ÁNGEL REGUEIRO-JANEIRO , AND FRANCISCO J. GONZÁLEZ-CASTAÑO
Information Technology Group, atlanTTic, School of Telecommunications Engineering, University of Vigo, Campus, 36310 Vigo, Spain
Corresponding author: Francisco J. González-Castaño (javier@gti.uvigo.es)
This work was supported in part by the Ministerio de Economía, Industria y Competitividad, Spain, under Grant TEC2016-76465-C2-2-R,
and in part by Xunta de Galicia under Grant GRC2018/053,ED341D-R2016/012.

ABSTRACT Micro-blogging sources such as the Twitter social network provide valuable real-time data for
market prediction models. Investors’ opinions in this network follow the fluctuations of the stock markets
and often include educated speculations on market opportunities that may have impact on the actions of
other investors. In view of this, we propose a novel system to detect positive predictions in tweets, a type
of financial emotions which we term ‘‘opportunities’’ that are akin to ‘‘anticipation’’ in Plutchik’s theory.
Specifically, we seek a high detection precision to present a financial operator a substantial amount of such
tweets while differentiating them from the rest of financial emotions in our system. We achieve it with
a three-layer stacked Machine Learning classification system with sophisticated features that result from
applying Natural Language Processing techniques to extract valuable linguistic information. Experimental
results on a dataset that has been manually annotated with financial emotion and ticker occurrence tags
demonstrate that our system yields satisfactory and competitive performance in financial opportunity
detection, with precision values up to 83%. This promising outcome endorses the usability of our system to
support investors’ decision making.

INDEX TERMS Financial management, machine learning, emotion recognition, natural language
processing.

I. INTRODUCTION sources. Two examples are Thomson Reuters Eikon,3 which

A. MOTIVATION
Among the many factors that compel people to take decisions
in stock markets, opinions on micro-blogging sites deserve
consideration [1]–[3]. It has been shown that fast paced infor-
mation in social media is valuable for sales prediction [4]–[6]
and has strong influence on micro-economic trends in stock
markets [7]. Investors’ decisions can be affected not only by
media content [1] but also by public mood [8]. Besides, the
advent of powerful social trading platforms, such as eToro1
and XTB online trading,2 has broadened the spectrum of
input data. There already exist online tools to analyse these
1 Available at https://www.etoro.com/, August 2020.
2 Available at https://www.xtb.com/, August 2020.
The associate editor coordinating the review of this manuscript and
approving it for publication was Kaustubh Raosaheb Patil
performs sentiment analysis ( SA) of social media content,
and the Bloomberg platform,4 which provides social media
indicators.
Furthermore, [9] argued that, even though stock mar-
kets contain enough information to define their state, this
information becomes obsolete before the general public
can process it to take decisions. Therefore, instantaneous
information in the Twitter micro-blogging network deserves
attention [10]–[12], and, in particular, in finance [7], [13].
Stock market data mining conforms a significant body of
research [14]–[18] supporting market predictability. Some
works have focused on Natural Language Processing (NLP)
techniques [19]–[21] and their application to emotion
analysis (EA) [22], [23].
3 Available at https://eikon.thomsonreuters.com/, August 2020.
. 4 Available at https://www.bloomberg.com/, August 2020.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 215679
 F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
To the best of our knowledge, this work is the first proposal
based on EA techniques to detect what we call financial oppor-
tunities, that is, Twitter posts that speculate or reason that
the value of particular actives will grow, rather than merely
expressing positive opinions about those actives. Specifically,
we present a three-layer stacked classifier with a first stage to
discard neutral entries, a second stage to distinguish generic
positive emotions from negative ones and a last stage to
differentiate opportunity entries –our ultimate goal– from any
other positive emotions.
For this purpose, the system is enhanced with sophisticated
linguistic information features such as n-gram sequences,
emotion and polarity dictionaries, and frequency counters of
hashtags, numerical information and percentages.
B. RESEARCH GOAL AND CONTRIBUTIONS
We seek a Machine Learning system to detect financial
opportunities in textual information extracted from tweets,
where by opportunities we refer to positive user speculations
and forecasts (e.g. ‘‘the average yield of Google is 15.5%’’)
instead of mere positive statements (e.g. ‘‘I like Google’’).
They are akin to looking forward positively for upcoming
events, as ‘‘anticipation’’ in Plutchik’s theory [24].
This problem has not been previously considered in aca-
demic research and has remarkable practical applications for
both stock market screening and decision making. Notwith-
standing this applied perspective, our research has also a
theoretical goal, an enhanced understanding of the linguistic
dimension of micro-blogging comments to extract investment
opportunities. Summing up, this work contributes to more
effective Machine Learning classifiers for stock market data
mining with the following:
• A three-layer stacked Machine Learning classification
system for detecting opportunities in micro-blogging
comments.
• An analysis of the most relevant features through NLP
techniques.
In addition, we employ a new dataset of investors’ com-
ments for stock market forecasting with 6,000 entries –similar
in size to those in other state-of-the-art studies [25]–[28],
which has been manually annotated with financial emotion
and ticker occurrence tags (e.g. in ‘‘the average yield of
Google is 15.5%’’ the ticker tag is GOOGL) by experts in
the field.
The rest of this article is organised as follows. Section II
reviews related work in knowledge extraction, discussing EA
from general and financial perspectives. Section III describes
the classification problem and our solution. Section IV
presents the experimental text corpus and the numerical tests
that validate our approach. Finally, Section V concludes this
article.
II. RELATED WORK
Novice investors in stock markets give high considera-
tion to the experience of financial experts. Thus, automatic
215680
knowledge extraction from the comments by experts in finan-
cial news, blogs and social media is an interesting research
goal and a valuable asset for practical applications [29], [30].
Previous researchers have analysed online sources to char-
acterise stock markets. Most of them have focused on news,
as [31], who studied the correlation between financial market
events and financial news content. In this regard, the work
by [32] deserves a special mention. Using Machine Learning
models of Latent Dirichlet Allocation, they concluded that
the information extracted from news sources is better than
price variations for predicting assets’ volatility. Additionally,
in [33] a Deep Learning ( DL) model for news categorisation
was presented proving that financial sources of information
have a significant effect on investments. There is also work on
knowledge extraction from financial blogs. For example, the
method by [34] combined stock indexes with sentiment time
series from micro-blogs. Besides, [29] applied multivariate
regression to financial blogs to study how the markets react
to the posts. Finally, the study of the information posted
on social media platforms like Twitter has been a relevant
research topic in recent years. Consider for example the
analysis of consumer profiles by [35] and the study of the evo-
lution of stock prices and social media content by [36] based
on a latent space model [37]. However, none of these works
has considered user speculations about the evolution of the
assets. Note that all of them have taken temporal information
directly from the timestamps of financial forecasts or posts
[29], [32], [34], [36]. In fact, we are not aware of any financial
research supported by temporal analysis at discursive level
such as ours.5
In this context, NLP techniques [40] such as SA [41] and
EA [42], [43] have successfully extracted knowledge from
comments of stock market experts. Specifically, in [42] the
authors presented a novel approach for the automatic extrac-
tion of a high-coverage and high-precision lexicon annotated
with emotion scores, while in [43] they confirmed the impact
of social media emotions on the stock market.
Often as an initial stage of EA, SA infers sentiment polarity
from language features of user opinions [41], [44]. Typically,
three polarity levels (negative, neutral and positive) are con-
sidered [45], although some authors add two deeper positive
and negative sentiment levels [46], [47]. Among the wealth
of research in Twitter mining, we can cite for example the
work by [48] on the analysis of consumer opinions about
commercial brands, or by [49] on financial SA. The latter
classified Twitter information about eight stock markets with
a Support Vector Classifier ( SVC).
EA infers feelings from linguistic information in user com-
ments [50], [51]. It has become a central topic in the field
of Affective Computing. Some typical emotion classes in
5 This aspect is relevant. Even though [38] already noted the interest
of the temporal perspective of the customers, i.e. the importance of their
own understanding of time, few works have addressed temporal analysis at
discursive level. Among early ones we must mention [39], which extracted
time from punctuation, word choices and academic terminology or keywords
in journal titles, by also considering common sense context information.
VOLUME 8, 2020
F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
the literature are love, joy, anger, sadness, fear and sur-
prise, as defined by [52]. For example, the approach by [53]
detected affection in virtual communication environments. EA
has numerous applications in education [54], healthcare [55],
[56] and gaming [57], to cite some. It has also caught the
attention of the industry as a profitable asset. In the particular
case of finance, [58] proposed a Linked Data approach for
modelling sentiment and emotions from tweets about secu-
rities of the Spanish stock market. The system for financial
forecasting by [22] incorporated different sources of informa-
tion into an integrated river model and combined GPOMS6 with
a Neural Network classifier, and [23] applied a multi-layer
perceptron to study the influence of investor emotions in
investment decisions. At the end, even though EA is still an
interesting research topic [59], [60], no previous analyses
have taken into consideration finance-related emotions, such
as opportunity or anticipation, and have focused exclusively
on general human emotions.
Training methodologies, on the other hand, can be divided
into supervised (manually annotated), semi-supervised and
unsupervised approaches. For example, in [61] some authors
of this article proposed an unsupervised system for auto-
matic generation of emotion lexica with polarity data. Hybrid
solutions such as [62] (which applied classifiers as well as
unsupervised approaches with polarity lexicons and syntactic
structures) have produced high competitive results. Finally,
stacking strategies combining different Machine Learning
techniques into the same predictive model can improve per-
formance [63], [64]. It has been our intent to exploit the ben-
efits of stacking in ensemble learning techniques to further
increase the performance of our system.
III. SYSTEM ARCHITECTURE
In this section we present our novel three-layer stacked sys-
tem for detecting financial tweets on market opportunities.
Specifically, the first stage detects neutral entries, the second
stage distinguishes between general positive and negative
emotions and the last stage separates opportunity entries from
any other positive statements. Figure 1 illustrates its frame-
work, which consists of two modules that will be described in
detail in the next subsections, where financial tweets are taken
from a training database. Table 1 shows examples of input
data and the corresponding emotion categories (including
‘‘opportunity’’, which we have already described). In the
context of Plutchik’s theory [24], ‘‘positive statement’’ cor-
responds to any positive emotion on present events such as
‘‘expectation’’; and ‘‘negative awareness’’ to any negative
opinion about an asset, regardless of the sentiment expressed
in the tweet, which is similar to ‘‘disapproval’’.
A. DATA PROCESSING MODULE
This module discards useless data. Specifically, we apply a
series of text transformations to ensure the quality of the data
6 Google Profile of Moods States (GPOMS). At the time this article was
written it was no longer available.
VOLUME 8, 2020
that enter the classification module. Table 2 presents some
examples before and after applying these transformations.
• Filtering. It selects finance-related tweets while
discarding spam and related tweets. Basically, tweets
containing asset identifiers such as stock markets’ tick-
ers, URLs and quantitative information. We also remove
advertising entries with spam-related words and expres-
sions such as sorteo ‘draw’, gratis ‘free’ and ?‘quieres
ganar dinero? ‘do you want to earn money?’. Moreover,
we discard words in other languages than Spanish using
the Enchant Python module.7 Finally, to discard highly
similar tweets, we implemented a similarity detector
based on the Jaccard distance with a threshold of 0.75%,
inspired by state-of-the-art spam detection techniques
[65]–[67].
• Hashtag and mention splitting. We decompose hashtags
in words with a splitter that uses our own lexica [68],
[69] and the Spanish frequency reference corpus (CREA)
by Real Academia Española de la Lengua.8 As shown
in the example in Table 2, the splitter divides the term
acuerdocomercial successfully as acuerdo comercial
‘commercial agreement’.
• Spelling correction. We correct words with spelling
mistakes by replacing them by the most likely candi-
date through our own word distance algorithm that uses
CREA as well as the Enchant Python module. In the
example in Table 2, the word precausion is corrected and
replaced by precaución ‘caution’.
• Stock market asset, mention and hashtag detection and
removal. We use regular expressions to detect capitalised
letters and identify representative symbols such as $, @
and #. Once identified, all these elements are removed
from the text of the entries.
• Stop words removal. Meaningless words such as deter-
miners and prepositions9 are removed from the text.
We also remove URLs and retweet ( RT) tags. We consider
days of the week and months of the year as stop words.
However, we keep elements such as no ‘not’, sí ‘yes’,
muy ‘very’ and poco ‘few’, since they help to interpret
the evolution of the assets.
• Quantitative data and laughter replacement. We sub-
stitute quantitative data (numbers and percentages) by
tags representing their sign, that is, ‘+’ and ‘−’ for
positive and negative figures, respectively. We decided
to use these symbolic representations rather than the
amounts themselves because they are more clearly
related to stock markets’ upward and downward move-
ments. Laughing onomatopoeic expressions such as
haha and hehe, which are typically composed of the
characters j or h interspersed with the vowels a, e and
i in any quantity, are replaced by the LAUGH tag.
7 Available at https://pypi.org/project/pyenchant/, August 2020.
8 Available at http://corpus.rae.es/lfrecuencias.html, August 2020.
9 Available at https://www.ranks.nl/stopwords/spanish, August 2020.
215681
 F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
FIGURE 1. Our proposed framework.
TABLE 1. Examples of dataset entries with emotion tags.
• Text lemmatisation. The content of the tweets is split into
tokens (words), which are independently checked in the
Hunspell dictionary.10 Finally, the words are lemmatised
using the Freeling NLP tool.11
B. CLASSIFICATION MODULE
1) MACHINE LEARNING MODEL: FEATURE PROCESSING
The features we selected to extract valuable information from
our dataset and build our model are:
• Char-grams, word tokens and word-grams. Char-grams
are sequences of adjacent characters in a text (note
that spaces must also be considered in this case). Word
tokens represent character grams only from the text
inside word boundaries, whereas word n-grams are
sequences of adjacent words in a text, among which
10 Available at https://hunspell.github.io/, August 2020.
11 Available at http://nlp.lsi.upc.edu/freeling/node/1, August 2020.
215682
we also include the textual representation of the emojis
in the tweets. To generate char n-grams, word tokens
and word n-grams we combined CountVectorizer12 with
GridSearchCV,13 both from the Scikit-Learn Python
library, using the parameters in Listing 1. GridSearchCV
performs an exhaustive search over specified parameter
ranges for an estimator. As a result, we obtained the
optimal parameters max_df = 0.5, min_df = 0.001 and
ngram_range = (1,7) in our scenario.
• Frequency counters. These features count the number
of hashtags, negative and positive amounts and per-
centages, exclamation marks, interrogation marks and
adverbs in the content of the tweets.
12 Available at https://scikit-learn.org/stable/ modules/generat-
ed/sklearn.feature_extraction.text.CountVectorizer.html, July 2020.
13 Available at https://scikit-learn.org/stable/ modules/generat-
ed/sklearn.model_selection.GridSearchCV.html, August 2020.
VOLUME 8, 2020
F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
TABLE 2. Examples of tweets before and after processing.
Listing. 1. Configuration for the generation of n-grams.
•Dictionary features. We use our polarity and emoji lexi-
con,14 for negative and positive sentiments (specifically
for characterisation of general negative and posi-
tive financial emotions), negative awareness, positive
statement and opportunity financial emotions; and an
external emotion lexicon,15 for sadness and happiness.
We count the words and emojis in the tweets correspond-
ing to each sentiment and emotion category of those
lexica.
• Temporal features. These features, based on the discur-
sive analysis in the preprocessing stage, are relevant
because speculative tweets (such as those expressing
caution or opportunities) tend to use certain verbal tenses
(such as conditional tenses) in conjunction with partic-
ular expressions like alto riesgo ‘high risk’ and ocasión
inmejorable ‘unbeatable occasion’. The temporal fea-
tures of a tweet count the verbs in past, present, future,
and conditional tenses.
Table 3 summarises the features to build our Machine
Learning model for financial opportunity detection.
14 Available at https://www.gti.uvigo.es/index.php/en /resources/8-
lexicon-of-polarity-and-list-of-emojis -by-polarity-and-emotion-for-
application -in-the-financial-field, August 2020.
15 Available at http://www.cic.ipn.mx/∼sidorov/SEL.txt, August 2020.
VOLUME 8, 2020
2) EMOTION ANALYSIS: THREE-LAYER STACKED SYSTEM
Figure 2 shows the flow diagram of the stacked system. It is
composed of three stages: a first stage that distinguishes
between neutral and non-neutral entries, a second stage to
distinguish between positive and negative emotions and a
last stage to extract opportunities from positive statements.
In each of these stages, we applied a decision depth thresh-
old to maximise opportunity detection, because in financial
analysis a high precision in key categories is preferable over
obtaining a large amount of positives [70].
The system includes implementations of Gradient Descent
(GD), Decision Tree (DT), Random Forest (RF) and Support
Vector Classification ( SVC) algorithms from the Scikit-Learn
Python library.16 Note that the size of our training datasets did
not justify the application of DL techniques [71], [72]. In this
regard, works such as [73], [74] are representative state of the
art examples of DL techniques for EA where the datasets are
much larger than ours.
Accordingly, we must characterise the level of confidence
(depth level) of emotion prediction precision. This can be
formalised with a non-parametric approach based on isotonic
regression [75]. This approach fits data with a non-decreasing
function defining a partial or total ordering, by minimising
quadratic expression (1) subject to (2).
X
αi (yi − ŷi )2 (1)
i
ŷmin = ŷ1 ≤ ŷ2 . . . ≤ ŷn = ŷmax (2)
In expression (1), each variable αi is a strictly positive
weight (1 by default), yi ∈ N is the true category of the i-th
16 Available at https://scikit-learn.org/stable/ supervised_learning.html#
supervised-learning, August 2020.
215683
 F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
TABLE 3. Input features in the machine learning model.
FIGURE 2. Flow diagram of our stacked classifier.
tweet and ŷi ∈ N its predicted category (so that yi and ŷi are
taken to a numerical space that allows defining a partial or
total ordering).
The level of confidence in the prediction depends on the
probabilities of correct class assignment. Logically, it is pos-
sible to select subsets of vectors in the training set for which
such probabilities are higher, and thus to trade accuracy by
precision.
IV. RESULTS AND DISCUSSION
All the experiments were executed on a computer with the
following specifications:
215684
• Operating System: Ubuntu 18.04.2 LTS 64 bits
• Processor: IntelCore i9-9900K 3.60 GHz
• RAM: 32GB DDR4
• Disk: 500 Gb (7200 rpm SATA) + 256 GB SSD
A. DATASET
TxStockData S.L., a fintech company, provided us with a
dataset composed of 6,000 tweets (similar in size to those
in other state-of-the-art studies [25]–[28]). The dataset was
gathered from 14th May 2019 to 3rd February 2020. Its
entries were manually annotated with financial emotion tags
by five experts in finance and NLP. At the end, the emotion
VOLUME 8, 2020
F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System

TABLE 4. Distribution of samples in the dataset by category. of the probability of presenting negative awareness results
as opportunities, A−
P+
, which would strongly discourage an
A−+
operator, that is τ2 = 1 − P
S ++ +P++ +NP+ +A−+
(where NP+
P P P
corresponds to neutral tweets from a financial perspective).
Therefore, our declared goal is first and foremost a reasonably
high precision in the detection of opportunities, for high
values of τ1 and, secondarily, of τ2 .

TABLE 5. Confusion matrix for tolerance calculations.

C. RESULTS
In this section, we evaluate our system to detect financial
opportunities. First we analyse the contributions the fea-
tures of the Machine Learning model (see Table 3) using
a basic single-layer classifier of opportunities vs. rest of
emotions (Section IV-C1). Then we present the precision
and tolerance results of a two-layer (neutral vs. non-neutral /
tag of each entry in the dataset was computed by majority opportunities vs. rest of emotions) stacked classifier, without
voting (in case of tie, it was selected randomly). To ensure the (Section IV-C2) and with (Section IV-C3) decision depth
quality of the data, we discarded repeated and spam entries, thresholds. Finally, we present our final three-layer stacked
yielding approximately 5,000 valid tweets.17 Each entry has classifier (which adds a final opportunities vs. positive state-
the following structure: ments layer) with decision depth thresholds, culminating
• ID: a unique numerical identifier. our incremental design (Section IV-C4). All results pre-
• Tweet: original text of the Tweet. sented in this section were computed by applying 10-fold
• Ticker: stock market assets mentioned in the text. cross-validation.
• Emotion: emotion label. Our model, with all characteristics, has ∼400,000 n-gram
Table 4 shows the distribution of the entries in the experi- features. This is computationally intractable. Consequently,
mental dataset by emotion category. we applied an attribute selector to extract the most relevant
features with a maximum decrease of 0.5% in precision.
B. DEFINITION OF PERFORMANCE METRICS Specifically, we chose the SelectPercentile18 method from
The practical goal of our system is to present to the user the Scikit-Learn Python library, as it outperformed other
the content of the tweets that are marked as opportunities. alternatives (SelectFromModel, SelectKBest, and RFECV).
Therefore, as previously said, we are interested in maximis- This method selects features according to a highest score
ing precision for the target financial emotion opportunity. percentile. We set a χ 2 score function and an 80th percentile
We defined two auxiliary tolerance metrics for those output threshold. At the end, we kept ∼50,000 n-gram features.
decisions that do not correspond to true opportunities but
will not necessarily discourage an operator, inspired by works 1) NUMERICAL TEST 1: RELEVANCE OF THE FEATURES
like [70]. Firstly we evaluated the performance of a single-layer classi-
We explain these tolerance metrics with the confusion fier that was trained to distinguish between opportunities and
matrix in Table 5. Notation XY represents the number all the other financial emotions in Table 4. Table 6 shows aver-
of tweets of emotion category X that are classified into age results for the target emotion opportunity with 10-fold
category Y . cross-validation. We started with a basic feature set only
From this confusion matrix we define tolerances τ1 and τ2 containing char-grams, word tokens and word-grams. Adding
as follows: all features in Section III-B1 was slightly advantageous for all
SP++ + P+ classifiers (up to ∼ 4% of improvement) except for the SVC.
P+
τ1 = (3) Moreover, in light of the results, RF seemed the best classifier
SP++ + P+
P+
+ NP+ + A−
P+ to detect opportunities, followed by the GD classifier.
SP++ + P+
P+
+ NP+ We considered these initial results (precision, τ1 and τ2 up
τ2 = (4)
SP++ + P+
P+
+ NP+ + A−
P+
to 64.33%, 70.65% and 87.31%, respectively) promising and
they motivated us to design more advanced approaches. The
In both cases the denominator is the sum of all decisions difference between τ1 and τ2 , which ranged between 15% and
marked as opportunities. Tolerance τ1 grows with SP+ +
, that is, 21%, indicated that a significant amount of neutral entries
when the user is presented positive statements as opportuni- were considered as opportunities. This is why we decided to
ties. Strictly these decisions are errors, but they have minimal
impact on user trust if P+P+  SP+ . Tolerance τ2 is the inverse
+

18 Available at https://scikit-learn.org/stable/modules /feature_selection.

17 We will make this dataset available to other researchers on request. html, August 2020.

VOLUME 8, 2020 215685

 F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
TABLE 6. Precisions and tolerances for numerical test 1.
TABLE 7. Precisions and tolerances of the two-layer stacked classifier,
numerical test 2.
explore the benefits of a stacked classifier with an initial stage
to filter neutral entries.
2) NUMERICAL TEST 2: TWO-LAYER STACKED CLASSIFIER
In this test we evaluated a two-layer stacked classifier. The
first stage was designed two distinguish between neutral and
non-neutral entries and the second stage distinguished oppor-
tunities from the rest of financial emotions in Table 4. In both
stages we used the same type of classifier but we applied inde-
pendent hyperparameter optimisation in each stage. Table 7
shows the resulting precisions and tolerances.
If we compare this test with numerical test 1, there was no
significant improvement with the GD and DT classifiers but
the stacked approach produced a 5% performance boost for
the other two. More specifically, the SVC and RF classifiers
produced the best results for all metrics and both scenarios
(basic set and all features, except for the precision of the
SVC if compared with the GD classifier with all features).
Furthermore, even though the GD and SVC classifiers behaved
similarly with all features, the GD classifier is far more com-
putationally intensive. DT precision results were considerably
bad, of around 40%. Thus we decided to discard GD and DT
for opportunity detection and keep only SVC and RF in further
analysis.
Ultimately, even though this test revealed further improve-
ment in precision and tolerance with the two-level stacked
classifier, the highest precision (with the RF algorithm) was
still under 75%. However, the tolerances against operator
discouragement were moderately satisfactory (τ1 >75%
and τ2 >85%).
215686
TABLE 8. Precisions and tolerances of the two-layer stacked classifier
with decision depth, numerical test 3.
3) NUMERICAL TEST 3: TWO-LAYER STACKED CLASSIFIER
WITH DECISION DEPTH
As previously stated, in businesses and investment, it is
preferable to obtain less positives for key categories with high
precision. Take personalised investment advertising as an
example. Banks must identify customer profiles with higher
success probability, since personalised commercialisation
actions are rather expensive. From a practical perspective,
we set the goal of detecting at least 10% of all opportunity
entries with high precision.
Accordingly, in this test we relied on the predict_
proba setting of the Scikit-Learn Python library to set class
thresholds. It allows computing the probabilities of the possi-
ble outcomes (classes in our model), thus providing levels of
confidence (see Section III-B2). When the probability of the
most likely class for a vector exceeded the threshold of that
class (which we term decision depth) the vector was assigned
that class. Otherwise, the entry was considered neutral. This
procedure was followed at both stages of numerical test 2,
with the same depth value, after empirical tests.
Table 8 shows the results of the two-layer stacked classifier
with decision depth. Both classifiers achieved significant
precision improvements. Besides, the RF classifier achieved a
precision above 70% in all cases for τ1 ∼ 80% and τ2 ∼ 90%.
Regardless of the fact that precision and τ1 were still not satis-
factory (they were less than 75% and 90%, respectively), the
values of τ2 for the two classifiers with all features indicated
that many negative awareness entries were still classified as
opportunities, which was unacceptable.
4) NUMERICAL TEST 4: THREE-LAYER STACKED CLASSIFIER
WITH DECISION DEPTH
All points considered, to meet our goals we finally applied the
three-layer stacked classifier framework (see Section III-B2)
with decision depth thresholds where the third stage distin-
guishes opportunities from positive statement entries.
We relied once again on the predict_proba setting of
the Scikit-Learn Python library to set the depth threshold.
As in the previous numerical test we used the same classifier
and depth threshold in all layers. Table 9 shows the results of
this test. Note the significant improvements by the SVC and RF
classifiers for all metrics. With the RF classifier we obtained
precision values above 80%, τ1 above 90% and τ2 above 95%,
meeting our initial requirements. Less than 5% of the negative
awareness entries were misclassified as opportunities, prob-
ably due to high ambiguity and, thus, annotation errors.
VOLUME 8, 2020
F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System

FIGURE 3. Histogram of ticker mentions in opportunities.

TABLE 9. Precisions and tolerances of the three-layer stacked classifier

with decision depth, numerical test 4.

TABLE 10. Improvements in precisions and tolerances between

numerical tests 1 and 4.

D. EXPLOITING THE RESULTS

For the sake of clarity and a handy comparison of the different
numerical tests, Table 10 shows the differences in precision,
τ1 and τ2 between numerical tests 1 and 4. The improvement
achieved with SVC and RF classifiers is noteworthy, supporting
the benefits of a stacking approach for financial opportunity
detection. The performance boost for RF, our final choice,
was almost 20% for precision and τ1 . In addition to meeting
our performance goals, we succeeded to identify as many
opportunity entries with high precision as our application
demanded, 9.27% and 10.43% on average with RF and SVC
respectively, with 10-fold cross-validation.
Moreover, Figure 3 represents the histogram of the men-
tions to assets in opportunity tweets. By examining the
specific assets that were mentioned between 14 May 2019
and 3 February 2020 in detected opportunities in the test-
ing sets in numerical experiment 4, we observed that those
assets corresponded to ∼ 80% of the mentions to the most
interesting stock actives (upper quartile of all mentions in
opportunities in the testing subsets, marked in red in Figure 3.
For clarity, only some representative well-known tickers are
shown in the x axis).
FIGURE 4. Possible integration of our system in a mobile app.
Therefore, a possible way to exploit these results would be
listing in a scroll section of the user dashboard the tweets we
pick as opportunities with the three-level stacked classifier.
Then we could highlight the tweets mentioning assets that
are detected following our approach with decision depth.
Figure 4 shows an example with real tweets as classified by
our system. Symbol is displayed next to the tweets that
contain opportunities that are detected with high precision.
V. CONCLUSION
Motivated by the strong influence of social media in users’
decisions in global markets, we propose a novel three-layer
stacked system to detect financial opportunities in tweets.
Our solution has been designed to extract such tweets with
high precision to present their content on personal finance

VOLUME 8, 2020 215687

 F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
management dashboards. It exploits sophisticated linguistic
features, such as polarity and emotion dictionaries and tem-
poral identification by discursive analysis, in its ML model.
It yields satisfactory and competitive market-level perfor-
mance in financial opportunity detection.
Experimental results, including two ad-hoc tolerance met-
rics focusing on data quality from an operator perspective,
demonstrate that our three-layer stacked system with decision
probability threshold (decision depth) succeeds to achieve our
goal. In particular, using a RF algorithm with all features in
the model, the system attains ∼ 83% financial opportunity
precision with detection tolerances τ1 ∼ 90% and τ2 ∼ 96%.
Given the valuable up-to-date information in micro-blogging
platforms, these promising results endorse the usability of our
system to support investors’ decision making.
As future work and to further expand the contributions
of our research, the emotion classifiers will be improved
with domain-specific tweet filters and quantitative (objec-
tive) numerical information from stock exchange prices.
In addition to opportunities, this will allow analysing positive
statement and negative awareness emotions in stock market
reports. Moreover, extensions for a multilingual version of
our framework to cover French and English will be provided.
REFERENCES
[1] T. Li, J. van Dalen, and P. J. van Rees, ‘‘More than just noise? Examining
the information content of stock microblogs on financial markets,’’ J. Inf.
Technol., vol. 33, no. 1, pp. 50–69, Mar. 2018.
[2] X. Li, P. Wu, and W. Wang, ‘‘Incorporating stock prices and news senti-
ments for stock market prediction: A case of Hong Kong,’’ Inf. Process.
Manage., vol. 57, no. 5, Sep. 2020, Art. no. 102212.
[3] G. Bello-Orgaz, R. M. Mesas, C. Zarco, V. Rodriguez, O. Cordón, and
D. Camacho, ‘‘Marketing analysis of wineries using social collective
behavior from users’ temporal activity on Twitter,’’ Inf. Process. Manage.,
vol. 57, no. 5, Sep. 2020, Art. no. 102220.
[4] M. Meire, M. Ballings, and D. Van den Poel, ‘‘The added value of social
media data in B2B customer acquisition systems: A real-life experiment,’’
Decis. Support Syst., vol. 104, pp. 26–37, Dec. 2017.
[5] P.-F. Pai and C.-H. Liu, ‘‘Predicting vehicle sales by sentiment anal-
ysis of Twitter data and stock market values,’’ IEEE Access, vol. 6,
pp. 57655–57662, 2018.
[6] H. Yuan, W. Xu, Q. Li, and R. Lau, ‘‘Topic sentiment mining for sales per-
formance prediction in e-commerce,’’ Ann. Oper. Res., vol. 270, nos. 1–2,
pp. 553–576, Nov. 2018.
[7] F. Mai, Z. Shan, Q. Bai, X. Wang, and R. H. L. Chiang, ‘‘How does social
media impact bitcoin value? A test of the silent majority hypothesis,’’
J. Manage. Inf. Syst., vol. 35, no. 1, pp. 19–52, Jan. 2018.
[8] Y. Sun, X. Liu, G. Chen, Y. Hao, and Z. J. Zhang, ‘‘How mood affects the
stock market: Empirical evidence from microblogs,’’ Inf. Manage., vol. 57,
no. 5, pp. 103–181, Jul. 2020.
[9] D. Enke and S. Thawornwong, ‘‘The use of data mining and neural
networks for forecasting stock market returns,’’ Expert Syst. Appl., vol. 29,
no. 4, pp. 927–940, Nov. 2005.
[10] M. S. Gerber, ‘‘Predicting crime using Twitter and kernel density estima-
tion,’’ Decis. Support Syst., vol. 61, pp. 115–125, May 2014.
[11] A. G. Reece, A. J. Reagan, K. L. M. Lix, P. S. Dodds, C. M. Danforth,
and E. J. Langer, ‘‘Forecasting the onset and course of mental illness with
Twitter data,’’ Sci. Rep., vol. 7, no. 1, pp. 1–11, Dec. 2017.
[12] K. Zahra, M. Imran, and F. O. Ostermann, ‘‘Automatic identification of
eyewitness messages on Twitter during disasters,’’ Inf. Process. Manage.,
vol. 57, no. 1, pp. 102–107, 2020.
[13] N. Oliveira, P. Cortez, and N. Areal, ‘‘The impact of microblogging data
for stock market prediction: Using Twitter to predict returns, volatility,
trading volume and survey sentiment indices,’’ Expert Syst. Appl., vol. 73,
pp. 125–144, May 2017.
215688
[14] M. Nofer and O. Hinz, ‘‘Using Twitter to predict the stock market,’’ Bus.
Inf. Syst. Eng., vol. 57, no. 4, pp. 229–242, Aug. 2015.
[15] T. Dimpfl and S. Jank, ‘‘Can Internet search queries help to predict stock
market volatility?’’ Eur. Financial Manage., vol. 22, no. 2, pp. 171–192,
Mar. 2016.
[16] X. Zhong and D. Enke, ‘‘Forecasting daily stock market return using
dimensionality reduction,’’ Expert Syst. Appl., vol. 67, pp. 126–139,
Jan. 2017.
[17] X. Zhang, Y. Zhang, S. Wang, Y. Yao, B. Fang, and P. S. Yu, ‘‘Improving
stock market prediction via heterogeneous information fusion,’’ Knowl.-
Based Syst., vol. 143, pp. 236–247, Mar. 2018.
[18] E. Hoseinzade and S. Haratizadeh, ‘‘CNNpred: CNN-based stock market
prediction using a diverse set of variables,’’ Expert Syst. Appl., vol. 129,
pp. 273–285, Sep. 2019.
[19] F. Sun, A. Belatreche, S. Coleman, T. M. McGinnity, and Y. Li, ‘‘Pre-
processing online financial text for sentiment classification: A natural
language processing approach,’’ in Proc. IEEE Conf. Comput. Intell.
Financial Eng. Econ. (CIFEr), Mar. 2014, pp. 122–129.
[20] I. E. Fisher, M. R. Garnsey, and M. E. Hughes, ‘‘Natural language process-
ing in accounting, auditing and finance: A synthesis of the literature with a
roadmap for future research,’’ Intell. Syst. Accounting, Finance Manage.,
vol. 23, no. 3, pp. 157–214, Jul. 2016.
[21] F. Z. Xing, E. Cambria, and R. E. Welsch, ‘‘Natural language based
financial forecasting: A survey,’’ Artif. Intell. Rev., vol. 50, no. 1, pp. 49–73,
Jun. 2018.
[22] K. K. Singh and P. Dimri, ‘‘Score based financial forecasting method by
incorporating different sources of information flow into integrative river
model,’’ in Proc. 6th Int. Conf. Cloud Syst. Big Data Eng. (Confluence),
Jan. 2016, pp. 685–688.
[23] N. I. M. Razi, M. Othman, and H. Yaacob, ‘‘Investment decisions
based on EEG emotion recognition,’’ Adv. Sci. Lett., vol. 23, no. 11,
pp. 11345–11349, Nov. 2017.
[24] R. Plutchik, ‘‘The circumplex as a general model of the structure of
emotions and personality,’’ in Circumplex Models of Personality and Emo-
tions. Worcester, MA, USA: American Psychological Association, 2004,
pp. 17–45.
[25] S. P. Chatzis, V. Siakoulis, A. Petropoulos, E. Stavroulakis, and
N. Vlachogiannakis, ‘‘Forecasting stock market crisis events using deep
and statistical machine learning techniques,’’ Expert Syst. Appl., vol. 112,
pp. 353–371, Dec. 2018.
[26] M. Al-Smadi, M. Al-Ayyoub, Y. Jararweh, and O. Qawasmeh, ‘‘Enhancing
aspect-based sentiment analysis of arabic Hotels’ reviews using morpho-
logical, syntactic and semantic features,’’ Inf. Process. Manage., vol. 56,
no. 2, pp. 308–319, Mar. 2019.
[27] D. Simester, A. Timoshenko, and S. I. Zoumpoulis, ‘‘Targeting prospec-
tive customers: Robustness of machine-learning methods to typical data
challenges,’’ Manage. Sci., vol. 66, no. 6, pp. 2495–2522, 2019.
[28] J. Tuke, A. Nguyen, M. Nasim, D. Mellor, A. Wickramasinghe, N. Bean,
and L. Mitchell, ‘‘Pachinko prediction: A Bayesian method for event pre-
diction from social media data,’’ Inf. Process. Manage., vol. 57, no. 2,
Mar. 2020, Art. no. 102147.
[29] L. K. Rickett, ‘‘Do financial blogs serve an infomediary role in capital
markets?’’ Amer. J. Bus., vol. 31, no. 1, pp. 17–40, Apr. 2016.
[30] W. He, F.-K. Wang, and V. Akula, ‘‘Managing extracted knowledge from
big social media data for business decision making,’’ J. Knowl. Manage.,
vol. 21, no. 2, pp. 275–294, Apr. 2017.
[31] M. Alanyali, H. S. Moat, and T. Preis, ‘‘Quantifying the relationship
between financial news and the stock market,’’ Sci. Rep., vol. 3, no. 1,
pp. 3578–3584, Dec. 2013.
[32] A. Atkins, M. Niranjan, and E. Gerding, ‘‘Financial news predicts stock
market volatility better than close price,’’ J. Finance Data Sci., vol. 4, no. 2,
pp. 120–137, Jun. 2018.
[33] M.-Y. Day and C.-C. Lee, ‘‘Deep learning for financial sentiment analysis
on finance news providers,’’ in Proc. IEEE/ACM Int. Conf. Adv. Social
Netw. Anal. Mining (ASONAM), Aug. 2016, pp. 1127–1134.
[34] Y. Wang, ‘‘Stock market forecasting with financial micro-blog based on
sentiment and time series analysis,’’ J. Shanghai Jiaotong Univ. Sci.,
vol. 22, no. 2, pp. 173–179, Apr. 2017.
[35] E. Ioanăs and I. Stoica, ‘‘Social media and its impact on consumers
behavior,’’ Int. J. Econ. Practices Theories, vol. 4, no. 2, pp. 295–303,
2014.
[36] A. Sun, M. Lachanski, and F. J. Fabozzi, ‘‘Trade the tweet: Social media
text mining and sparse matrix factorization for stock market prediction,’’
Int. Rev. Financial Anal., vol. 48, pp. 272–281, Dec. 2016.
VOLUME 8, 2020
F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System
[37] F. Ming, F. Wong, Z. Liu, and M. Chiang, ‘‘Stock market prediction from
WSJ: Text mining via sparse matrix factorization,’’ in Proc. IEEE Int. Conf.
Data Mining, Dec. 2014, pp. 430–439.
[38] P. T. Gibbs, ‘‘Time, temporality and consumer behaviour,’’ Eur. J. Market-
ing, vol. 32, nos. 11–12, pp. 993–1007, Dec. 1998.
[39] J. M. Forray and J. Woodilla, ‘‘Artefacts of management academe,’’ Time
Soc., vol. 14, nos. 2–3, pp. 323–339, Sep. 2005.
[40] B. Liu, Sentiment Analysis: Mining Opinions, Sentiments, and Emotions.
Cambridge, U.K.: Cambridge Univ. Press, 2015.
[41] A. Derakhshan and H. Beigy, ‘‘Sentiment analysis on stock social media
for stock price movement prediction,’’ Eng. Appl. Artif. Intell., vol. 85,
pp. 569–578, Oct. 2019.
[42] J. Staiano and M. Guerini, ‘‘Depeche mood: A lexicon for emotion analysis
from crowd annotated news,’’ in Proc. 52nd Annu. Meeting Assoc. Comput.
Linguistics, vol. 2. Stroudsburg, PA, USA: Association for Computational
Linguistics, 2014, pp. 427–433.
[43] Y. Ge, J. Qiu, Z. Liu, W. Gu, and L. Xu, ‘‘Beyond negative and positive:
Exploring the effects of emotions in social media during the stock market
crash,’’ Inf. Process. Manage., vol. 57, no. 4, Jul. 2020, Art. no. 102218.
[44] G. Xu, Y. Meng, X. Qiu, Z. Yu, and X. Wu, ‘‘Sentiment analysis of
comment texts based on BiLSTM,’’ IEEE Access, vol. 7, pp. 51522–51532,
2019.
[45] Y. Fang, H. Tan, and J. Zhang, ‘‘Multi-strategy sentiment analysis of
consumer reviews based on semantic fuzziness,’’ IEEE Access, vol. 6,
pp. 20625–20631, 2018.
[46] A. Balahur and J. M. Perea-Ortega, ‘‘Sentiment analysis system adaptation
for multilingual processing: The case of tweets,’’ Inf. Process. Manage.,
vol. 51, no. 4, pp. 547–556, Jul. 2015.
[47] J. Bučar, J. Povh, and M. Žnidaršič, ‘‘Sentiment classification of the
Slovenian news texts,’’ in Proc. 9th Int. Conf. Comput. Recognit. Syst.
Cham, Switzerland: Springer, 2016, pp. 777–787.
[48] D. Zimbra, M. Ghiassi, and S. Lee, ‘‘Brand-related Twitter sentiment anal-
ysis using feature engineering and the dynamic architecture for artificial
neural networks,’’ in Proc. 49th Hawaii Int. Conf. Syst. Sci. (HICSS),
Jan. 2016, pp. 1930–1938.
[49] J. Smailović, M. Grčar, N. Lavrač, and M. Žnidaršič, ‘‘Predictive sentiment
analysis of tweets: A stock market application,’’ in Human-Computer
Interaction and Knowledge Discovery in Complex, Unstructured, Big
Data (Lecture Notes in Computer Science), vol. 7947. Berlin, Germany:
Springer, 2013, pp. 77–88.
[50] J. K. Rout, K.-K.-R. Choo, A. K. Dash, S. Bakshi, S. K. Jena, and
K. L. Williams, ‘‘A model for sentiment and emotion analysis of unstruc-
tured social media text,’’ Electron. Commerce Res., vol. 18, no. 1,
pp. 181–199, Mar. 2018.
[51] C.-H. Chen, W.-P. Lee, and J.-Y. Huang, ‘‘Tracking and recognizing emo-
tions in short text messages from online chatting services,’’ Inf. Process.
Manage., vol. 54, no. 6, pp. 1325–1344, Nov. 2018.
[52] W. G. Parrott, Emotions in Social Psychology: Essential Readings. Hove,
U.K.: Psychology Press, 2001.
[53] A. Neviarouskaya, H. Prendinger, and M. Ishizuka, ‘‘Textual affect sens-
ing for sociable and expressive online communication,’’ in Proc. Int.
Conf. Affect. Comput. Intell. Interact. Berlin, Germany: Springer, 2007,
pp. 218–229.
[54] J. Xu, Z. Huang, M. Shi, and M. Jiang, ‘‘Emotion detection in E-learning
using expectation-maximization deep spatial-temporal inference net-
work,’’ in Advances in Intelligent Systems and Computing, vol. 650. Cham,
Switzerland: Springer, 2018, pp. 245–252.
[55] M. Z. Asghar, A. Khan, K. Khan, H. Ahmad, and I. A. Khan, ‘‘COGEMO:
Cognitive-based emotion detection from patient generated health reviews,’’
J. Med. Imag. Health Informat., vol. 7, no. 6, pp. 1436–1444, Oct. 2017.
[56] S. Z. Bong, K. Wan, M. Murugappan, N. M. Ibrahim, Y. Rajamanickam,
and K. Mohamad, ‘‘Implementation of wavelet packet transform and non
linear analysis for emotion classification in stroke patient using brain
signals,’’ Biomed. Signal Process. Control, vol. 36, pp. 102–112, Jul. 2017.
[57] M. S. Hossain, G. Muhammad, B. Song, M. M. Hassan, A. Alelaiwi,
and A. Alamri, ‘‘Audio–visual emotion-aware cloud gaming framework,’’
IEEE Trans. Circuits Syst. Video Technol., vol. 25, no. 12, pp. 2105–2118,
Dec. 2015.
[58] J. F. Sánchez-Rada, M. Torres, C. A. Iglesias, R. Maestre, and E. Peinado,
‘‘A linked data approach to sentiment and emotion analysis of Twitter in the
financial domain,’’ in Proc. 2nd Int. Workshop Semantic Web Enterprise
Adoption Best Pract., 2nd Int. Workshop Finance Econ. Semantic Web,
vol. 1240, 2014, pp. 51–62.
VOLUME 8, 2020
[59] D. Duxbury, T. Gärling, A. Gamble, and V. Klass, ‘‘How emotions
influence behavior in financial markets: A conceptual analysis and
emotion-based account of buy-sell preferences,’’ Eur. J. Finance, vol. 26,
pp. 1–22, Mar. 2020.
[60] S. Pengnate and F. J. Riggins, ‘‘The role of emotion in P2P microfinance
funding: A sentiment analysis approach,’’ Int. J. Inf. Manage., vol. 54,
Oct. 2020, Art. no. 102138.
[61] M. Fernández-Gavilanes, J. Juncal-Martínez, S. García-Méndez,
E. Costa-Montenegro, and F. J. González-Castaño, ‘‘Creating Emoji
Lexica from unsupervised sentiment analysis of their descriptions,’’
Expert Syst. Appl., vol. 103, pp. 74–91, Aug. 2018.
[62] M. Fernández-Gavilanes, T. Àlvarez-López, J. Juncal-Martínez,
E. Costa-Montenegro, and F. J. González-Castaño, ‘‘GTI: An unsupervised
approach for sentiment analysis in Twitter,’’ in Proc. 9th Int. Workshop
Semantic Eval. (SemEval), 2015, pp. 35–40.
[63] A. Mehmood, B.-W. On, I. Lee, I. Ashraf, and G. Sang Choi, ‘‘Spam
comments prediction using stacking with ensemble learning,’’ J. Phys.,
Conf. Ser., vol. 933, Jan. 2018, Art. no. 012012.
[64] Y. Wang, S. Liu, S. Li, J. Duan, Z. Hou, J. Yu, and K. Ma, ‘‘Stacking-based
ensemble learning of self-media data for marketing intention detection,’’
Future Internet, vol. 11, no. 7, p. 155, Jul. 2019.
[65] S. R. Harsule and M. K. Nighot, ‘‘N-gram classifier system to filter spam
messages from OSN user wall,’’ in Advances in Intelligent Systems and
Computing. Singapore: Springer, 2016, pp. 21–28.
[66] S. Bajaj, N. Garg, and S. K. Singh, ‘‘A novel user-based spam review
detection,’’ Procedia Comput. Sci., vol. 122, pp. 1009–1015, Jan. 2017.
[67] S. Temma, M. Sugii, and H. Matsuno, ‘‘The document similarity index
based on the jaccard distance for mail filtering,’’ in Proc. 34th Int. Tech.
Conf. Circuits/Syst., Comput. Commun. (ITC-CSCC), Jun. 2019, pp. 1–4.
[68] S. García-Méndez, M. Fernández-Gavilanes, E. Costa-Montenegro,
J. Juncal-Martínez, and F. J. González-Castaño, ‘‘Automatic natural
language generation applied to alternative and augmentative
communication for online video content services using SimpleNLG
for Spanish,’’ in Proc. Internet Accessible Things, Apr. 2018, pp. 1–4.
[69] S. García-Méndez, M. Fernández-Gavilanes, E. Costa-Montenegro,
J. Juncal-Martínez, and F. Javier González-Castaño, ‘‘A library for
automatic natural language generation of spanish texts,’’ Expert Syst.
Appl., vol. 120, pp. 372–386, Apr. 2019.
[70] J. Jurgovsky, M. Granitzer, K. Ziegler, S. Calabretto, P.-E. Portier,
L. He-Guelton, and O. Caelen, ‘‘Sequence classification for credit-card
fraud detection,’’ Expert Syst. Appl., vol. 100, pp. 234–245, Jun. 2018.
[71] N. Papernot, P. McDaniel, S. Jha, M. Fredrikson, Z. B. Celik, and
A. Swami, ‘‘The limitations of deep learning in adversarial settings,’’ in
Proc. IEEE Eur. Symp. Secur. Privacy (EuroS P), Mar. 2016, pp. 372–387.
[72] R. Keshari, S. Ghosh, S. Chhabra, M. Vatsa, and R. Singh, ‘‘Unravelling
small sample size problems in the deep learning world,’’ in Proc. IEEE 6th
Int. Conf. Multimedia Big Data (BigMM), Sep. 2020, pp. 134–143.
[73] M. Abdul-Mageed and L. Ungar, ‘‘EmoNet: Fine-grained emotion detec-
tion with gated recurrent neural networks,’’ in Proc. 55th Annu. Meeting
Assoc. Comput. Linguistics. Stroudsburg, PA, USA: Association for Com-
putational Linguistics, 2017, pp. 718–728.
[74] L. Santamaria-Granados, M. Munoz-Organero, G. Ramirez-Gonzalez,
E. Abdulhay, and N. Arunkumar, ‘‘Using deep convolutional neural net-
work for emotion detection on a physiological signals dataset (AMIGOS),’’
IEEE Access, vol. 7, pp. 57–67, 2019.
[75] R. Luss and S. Rosset, ‘‘Generalized isotonic regression,’’ J. Comput.
Graph. Statist., vol. 23, no. 1, pp. 192–210, Jan. 2014.
FRANCISCO DE ARRIBA-PÉREZ received the
B.S. degree in telecommunication technologies
engineering, the M.S. degree in telecommunica-
tion engineering, and the Ph.D. degree from the
University of Vigo, Spain, in 2013, 2014, and
2019, respectively.
He is currently a Researcher with the Infor-
mation Technologies Group, University of Vigo,
Spain. His research includes the development of
machine learning solutions for different use cases
among the ones financial and heath must be highlighted.
215689
 F. de Arriba-Pérez et al.: Detection of Financial Opportunities in Micro-Blogging Data With a Stacked Classification System

SILVIA GARCÍA-MÉNDEZ received the B.S. FRANCISCO J. GONZÁLEZ-CASTAÑO received

degree in telecommunication technologies engi-
neering and the M.S. degree in telecommunication
engineering from the University of Vigo, Spain,
in 2014 and 2016, respectively.
She is currently pursuing the Ph.D. degree in
information and communications technology from
the University of Vigo, Spain. Since 2015, she
has been a Researcher with the Information Tech-
nologies Group, University of Vigo, Spain. Her
research includes the development of automatic solutions for natural lan-
guage processing in Spanish and English.
Ms. García-Méndez awards and honors include the Connecting for Good
Vodafone Award in 2017.
the B.S. degree from the University of Santi-
ago de Compostela, Spain, in 1990, and the
Ph.D. degree from the University of Vigo, Spain,
in 1998.
He is currently a Professor with the Univer-
sity of Vigo, Spain, where he leads the Infor-
mation Technologies Group. He has authored
over 100 articles in international journals, in the
fields of telecommunications and computer sci-
ence. He has participated in several relevant national and international
projects. He holds three U.S. patents.

JOSÉ ÁNGEL REGUEIRO-JANEIRO received

the B.S. degree in telecommunication technologies
engineering and the M.S. degree in telecommu-
nication engineering from the University of Vigo,
Spain, in 2017 and 2019, respectively.
He is currently pursuing the Ph.D. degree in
information and communications technology with
the University of Vigo, Spain. Since 2019, he has
been a Researcher with the Information Tech-
nologies Group, University of Vigo. His research
includes the development of machine learning solutions for the financial
field.

215690 VOLUME 8, 2020